Correlation between Sensing Accuracy and Read Margin of a Memristor-Based NO Gas Sensor Array Estimated by Neural...

Area of Science:

• Materials Science
• Sensor Technology
• Artificial Intelligence

Background:

• Memristor-based gas sensors (gasistors) show promise for NO gas detection and neural network (NN) analysis.
• Single gasistor use in NNs can lead to overfitting, degrading prediction accuracy.

Purpose of the Study:

• To propose a metal-insulator-silicon (MIS)-structured Zr3N4-based gasistor array to overcome NN overfitting issues.
• To enhance both the accuracy of NN analysis and the operating power efficiency for NO gas detection.

Main Methods:

• Development of a Zr3N4-based gasistor array with a metal-insulator-silicon (MIS) structure.
• Comparison of the array's performance against single metal/insulator/metal (MIM) and MIS gas sensor cells.

Main Results:

• The proposed MIS gasistor array demonstrated a decrease in training epochs.
• Achieved a 2.5% improvement in prediction accuracy at room temperature compared to single-cell sensors.
• The array effectively mitigated overfitting issues inherent in single-sensor NN analyses.

Conclusions:

• An array structure based on MIS can efficiently solve the overfitting issue in NN analysis for gas sensing.
• The Zr3N4-based MIS gasistor array offers improved accuracy and power efficiency for NO gas detection.
• This approach provides a robust solution for reliable gas sensing applications using neural networks.